Image Sensor Neural Network Processing Structure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image signal processing units face challenges with high power consumption due to frame-based imaging, inefficient processing of both detected and undetected photons, and the need for extensive data transmission and processing components.
Innovation Solution
An image sensor with a neural network processing structure where each photo detector is directly connected to the processing means, eliminating the need for conventional processors, analog-to-digital converters, and random access memory, and allowing for massively parallel data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If frame-based imaging is used with traditional processors, then image processing can be performed, but power consumption increases significantly
Solution Approach 1:
The patent extracts and eliminates unnecessary processing components (ADCs, RAM, conventional processors) from the traditional imaging system, keeping only the essential photo detectors directly connected to processing elements. This extraction removes the sources of high power consumption while maintaining core imaging functionality.
Solution Approach 2:
The patent replaces the mechanical/electrical processing system (conventional processors, ADCs, RAM) with a neural network-based processing system. This substitution enables parallel processing of photon detections without the power overhead of traditional sequential processing components.
2Use of energy by stationary object
If all photo detector outputs are processed regardless of detection status, then complete data processing is achieved, but unnecessary power consumption occurs
Solution Approach 1:
The patent applies partial action by processing only the necessary subset of photo detector outputs - specifically, only those detectors that actually detected photons. The neural network processes detections selectively rather than forcing processing of all detectors, eliminating wasted energy on undetected pixels while maintaining complete information about actual detections.
3Measurement precision
If high resolution quantizers and ADCs are used, then measurement precision improves, but device complexity and power consumption increase
Solution Approach 1:
The patent extracts and removes ADCs and high-resolution quantizers from the system. Instead, it uses photo detectors that directly output logic signals (0 or 1) representing photon detections. This extraction eliminates complex conversion components while maintaining sufficient precision for detecting the presence or absence of photons.
Solution Approach 2:
The patent replaces expensive, complex ADCs and high-resolution quantizers with simple, low-cost logic signal outputs from photo detectors. These simple binary outputs are sufficient for the detection task and eliminate the need for complex conversion hardware.
4Loss of energy
If data busses and transmission components are added, then data transmission capability improves, but power loss and system inefficiency increase
Solution Approach 1:
The patent merges the photo detectors with the processing means by placing processing elements directly adjacent to each photo detector and connecting them with minimal wiring. This merging eliminates separate data busses and transmission components, reducing power loss while maintaining full data transmission capability between detectors and processors.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This solution significantly reduces power consumption and increases processing speed while achieving a compact and efficient optical sensing system that only processes detected photons, thereby overcoming limitations of frame-based imaging.
Implementation Method 1
a photo detector adapted to output a logic signal
Data Source
AI summary
The present invention relates to an image sensor (1) for efficient optical sensing. The sensor (1) comprises a plurality of pixel sensors (3), each pixel sensor (3) comprising a photo detector (4), each pixel being adapted to output a digital signal. The sensor (1) further comprises processing means. for example a neural network (5), comprising a plurality of neurons (6). Each photo detector (4) is in the vicinity of said processing means. Furthermore, each photo detector (4) is connected to the processing means.